X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=3c2697dec3edb71740cf51d3eddd9cd7fe7b2828;hp=101024d10c24b4ec34c90693cd0244a4f2feb7a6;hb=13c702188833e778c6461c3c8d64be1d159f16ee;hpb=3e260a46ea9a8de224c3d0a29a608da3440f284a diff --git a/backend.c b/backend.c index 101024d1..3c2697de 100644 --- a/backend.c +++ b/backend.c @@ -52,11 +52,15 @@ #include "server.h" #include "lib/getrusage.h" #include "idletime.h" +#include "err.h" +#include "lib/tp.h" + +static pthread_t helper_thread; +static pthread_mutex_t helper_lock; +pthread_cond_t helper_cond; +int helper_do_stat = 0; -static pthread_t disk_util_thread; -static struct fio_mutex *disk_thread_mutex; static struct fio_mutex *startup_mutex; -static struct fio_mutex *writeout_mutex; static struct flist_head *cgroup_list; static char *cgroup_mnt; static int exit_value; @@ -72,7 +76,7 @@ unsigned int stat_number = 0; int shm_id = 0; int temp_stall_ts; unsigned long done_secs = 0; -volatile int disk_util_exit = 0; +volatile int helper_exit = 0; #define PAGE_ALIGN(buf) \ (char *) (((uintptr_t) (buf) + page_mask) & ~page_mask) @@ -86,7 +90,7 @@ static void sig_int(int sig) fio_server_got_signal(sig); else { log_info("\nfio: terminating on signal %d\n", sig); - fflush(stdout); + log_info_flush(); exit_value = 128; } @@ -182,7 +186,11 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, ratemin); return 1; } else { - rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; + if (spent) + rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; + else + rate = 0; + if (rate < ratemin || bytes < td->rate_bytes[ddir]) { log_err("%s: min rate %u not met, got" @@ -200,7 +208,11 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now, td->o.name, rate_iops); return 1; } else { - rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; + if (spent) + rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; + else + rate = 0; + if (rate < rate_iops_min || iops < td->rate_blocks[ddir]) { log_err("%s: min iops rate %u not met," @@ -345,7 +357,7 @@ static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) return 0; if (!td->o.timeout) return 0; - if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) + if (utime_since(&td->epoch, t) >= td->o.timeout) return 1; return 0; @@ -382,7 +394,7 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, * fill_device option is set. */ td_clear_error(td); - td->terminate = 1; + fio_mark_td_terminate(td); return 1; } else { /* @@ -451,7 +463,7 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) if (runtime_exceeded(td, &td->tv_cache)) { __update_tv_cache(td); if (runtime_exceeded(td, &td->tv_cache)) { - td->terminate = 1; + fio_mark_td_terminate(td); break; } } @@ -478,6 +490,12 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) break; while ((io_u = get_io_u(td)) != NULL) { + if (IS_ERR(io_u)) { + io_u = NULL; + ret = FIO_Q_BUSY; + goto reap; + } + /* * We are only interested in the places where * we wrote or trimmed IOs. Turn those into @@ -508,6 +526,11 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes) break; } + if (verify_state_should_stop(td, io_u)) { + put_io_u(td, io_u); + break; + } + if (td->o.verify_async) io_u->end_io = verify_io_u_async; else @@ -574,6 +597,7 @@ sync_done: * completed io_u's first. Note that we can get BUSY even * without IO queued, if the system is resource starved. */ +reap: full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); if (full || !td->o.iodepth_batch_complete) { min_events = min(td->o.iodepth_batch_complete, @@ -615,9 +639,22 @@ sync_done: dprint(FD_VERIFY, "exiting loop\n"); } +static unsigned int exceeds_number_ios(struct thread_data *td) +{ + unsigned long long number_ios; + + if (!td->o.number_ios) + return 0; + + number_ios = ddir_rw_sum(td->this_io_blocks); + number_ios += td->io_u_queued + td->io_u_in_flight; + + return number_ios >= td->o.number_ios; +} + static int io_bytes_exceeded(struct thread_data *td) { - unsigned long long bytes; + unsigned long long bytes, limit; if (td_rw(td)) bytes = td->this_io_bytes[DDIR_READ] + td->this_io_bytes[DDIR_WRITE]; @@ -628,7 +665,12 @@ static int io_bytes_exceeded(struct thread_data *td) else bytes = td->this_io_bytes[DDIR_TRIM]; - return bytes >= td->o.size; + if (td->o.io_limit) + limit = td->o.io_limit; + else + limit = td->o.size; + + return bytes >= limit || exceeds_number_ios(td); } /* @@ -642,7 +684,7 @@ static uint64_t do_io(struct thread_data *td) uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; unsigned int i; int ret = 0; - uint64_t bytes_issued = 0; + uint64_t total_bytes, bytes_issued = 0; if (in_ramp_time(td)) td_set_runstate(td, TD_RAMP); @@ -651,6 +693,16 @@ static uint64_t do_io(struct thread_data *td) lat_target_init(td); + /* + * If verify_backlog is enabled, we'll run the verify in this + * handler as well. For that case, we may need up to twice the + * amount of bytes. + */ + total_bytes = td->o.size; + if (td->o.verify != VERIFY_NONE && + (td_write(td) && td->o.verify_backlog)) + total_bytes += td->o.size; + while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) || td->o.time_based) { @@ -670,7 +722,7 @@ static uint64_t do_io(struct thread_data *td) if (runtime_exceeded(td, &td->tv_cache)) { __update_tv_cache(td); if (runtime_exceeded(td, &td->tv_cache)) { - td->terminate = 1; + fio_mark_td_terminate(td); break; } } @@ -678,11 +730,18 @@ static uint64_t do_io(struct thread_data *td) if (flow_threshold_exceeded(td)) continue; - if (bytes_issued >= (uint64_t) td->o.size) + if (bytes_issued >= total_bytes) break; io_u = get_io_u(td); - if (!io_u) { + if (IS_ERR_OR_NULL(io_u)) { + int err = PTR_ERR(io_u); + + io_u = NULL; + if (err == -EBUSY) { + ret = FIO_Q_BUSY; + goto reap; + } if (td->o.latency_target) goto reap; break; @@ -697,6 +756,18 @@ static uint64_t do_io(struct thread_data *td) */ if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { + + if (!td->o.verify_pattern_bytes) { + io_u->rand_seed = __rand(&td->verify_state); + if (sizeof(int) != sizeof(long *)) + io_u->rand_seed *= __rand(&td->verify_state); + } + + if (verify_state_should_stop(td, io_u)) { + put_io_u(td, io_u); + break; + } + if (td->o.verify_async) io_u->end_io = verify_io_u_async; else @@ -707,21 +778,37 @@ static uint64_t do_io(struct thread_data *td) else td_set_runstate(td, TD_RUNNING); + /* + * Always log IO before it's issued, so we know the specific + * order of it. The logged unit will track when the IO has + * completed. + */ + if (td_write(td) && io_u->ddir == DDIR_WRITE && + td->o.do_verify && + td->o.verify != VERIFY_NONE && + !td->o.experimental_verify) + log_io_piece(td, io_u); + ret = td_io_queue(td, io_u); switch (ret) { case FIO_Q_COMPLETED: if (io_u->error) { ret = -io_u->error; + unlog_io_piece(td, io_u); clear_io_u(td, io_u); } else if (io_u->resid) { int bytes = io_u->xfer_buflen - io_u->resid; struct fio_file *f = io_u->file; bytes_issued += bytes; + + trim_io_piece(td, io_u); + /* * zero read, fail */ if (!bytes) { + unlog_io_piece(td, io_u); td_verror(td, EIO, "full resid"); put_io_u(td, io_u); break; @@ -762,6 +849,7 @@ sync_done: bytes_issued += io_u->xfer_buflen; break; case FIO_Q_BUSY: + unlog_io_piece(td, io_u); requeue_io_u(td, &io_u); ret2 = td_io_commit(td); if (ret2 < 0) @@ -847,7 +935,7 @@ reap: if (td->o.fill_device && td->error == ENOSPC) { td->error = 0; - td->terminate = 1; + fio_mark_td_terminate(td); } if (!td->error) { struct fio_file *f; @@ -899,6 +987,9 @@ static void cleanup_io_u(struct thread_data *td) io_u_rexit(&td->io_u_requeues); io_u_qexit(&td->io_u_freelist); io_u_qexit(&td->io_u_all); + + if (td->last_write_comp) + sfree(td->last_write_comp); } static int init_io_u(struct thread_data *td) @@ -989,7 +1080,7 @@ static int init_io_u(struct thread_data *td) * Fill the buffer with the pattern if we are * going to be doing writes. */ - fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0); + fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); } } @@ -1015,6 +1106,14 @@ static int init_io_u(struct thread_data *td) p += max_bs; } + if (td->o.verify != VERIFY_NONE) { + td->last_write_comp = scalloc(max_units, sizeof(uint64_t)); + if (!td->last_write_comp) { + log_err("fio: failed to alloc write comp data\n"); + return 1; + } + } + return 0; } @@ -1055,12 +1154,14 @@ static int switch_ioscheduler(struct thread_data *td) /* * Read back and check that the selected scheduler is now the default. */ - ret = fread(tmp, 1, sizeof(tmp), f); + ret = fread(tmp, sizeof(tmp), 1, f); if (ferror(f) || ret < 0) { td_verror(td, errno, "fread"); fclose(f); return 1; } + tmp[sizeof(tmp) - 1] = '\0'; + sprintf(tmp2, "[%s]", td->o.ioscheduler); if (!strstr(tmp, tmp2)) { @@ -1076,6 +1177,8 @@ static int switch_ioscheduler(struct thread_data *td) static int keep_running(struct thread_data *td) { + unsigned long long limit; + if (td->done) return 0; if (td->o.time_based) @@ -1084,18 +1187,28 @@ static int keep_running(struct thread_data *td) td->o.loops--; return 1; } + if (exceeds_number_ios(td)) + return 0; - if (td->o.size != -1ULL && ddir_rw_sum(td->io_bytes) < td->o.size) { + if (td->o.io_limit) + limit = td->o.io_limit; + else + limit = td->o.size; + + if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) { uint64_t diff; /* * If the difference is less than the minimum IO size, we * are done. */ - diff = td->o.size - ddir_rw_sum(td->io_bytes); + diff = limit - ddir_rw_sum(td->io_bytes); if (diff < td_max_bs(td)) return 0; + if (fio_files_done(td)) + return 0; + return 1; } @@ -1119,6 +1232,50 @@ static int exec_string(struct thread_options *o, const char *string, const char return ret; } +/* + * Dry run to compute correct state of numberio for verification. + */ +static uint64_t do_dry_run(struct thread_data *td) +{ + uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; + + td_set_runstate(td, TD_RUNNING); + + while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || + (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td)) { + struct io_u *io_u; + int ret; + + if (td->terminate || td->done) + break; + + io_u = get_io_u(td); + if (!io_u) + break; + + io_u->flags |= IO_U_F_FLIGHT; + io_u->error = 0; + io_u->resid = 0; + if (ddir_rw(acct_ddir(io_u))) + td->io_issues[acct_ddir(io_u)]++; + if (ddir_rw(io_u->ddir)) { + io_u_mark_depth(td, 1); + td->ts.total_io_u[io_u->ddir]++; + } + + if (td_write(td) && io_u->ddir == DDIR_WRITE && + td->o.do_verify && + td->o.verify != VERIFY_NONE && + !td->o.experimental_verify) + log_io_piece(td, io_u); + + ret = io_u_sync_complete(td, io_u, bytes_done); + (void) ret; + } + + return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM]; +} + /* * Entry point for the thread based jobs. The process based jobs end up * here as well, after a little setup. @@ -1138,11 +1295,6 @@ static void *thread_main(void *data) } else td->pid = gettid(); - /* - * fio_time_init() may not have been called yet if running as a server - */ - fio_time_init(); - fio_local_clock_init(o->use_thread); dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); @@ -1169,13 +1321,6 @@ static void *thread_main(void *data) fio_mutex_down(td->mutex); dprint(FD_MUTEX, "done waiting on td->mutex\n"); - /* - * the ->mutex mutex is now no longer used, close it to avoid - * eating a file descriptor - */ - fio_mutex_remove(td->mutex); - td->mutex = NULL; - /* * A new gid requires privilege, so we need to do this before setting * the uid. @@ -1201,6 +1346,15 @@ static void *thread_main(void *data) * allocations. */ if (o->cpumask_set) { + if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) { + ret = fio_cpus_split(&o->cpumask, td->thread_number - 1); + if (!ret) { + log_err("fio: no CPUs set\n"); + log_err("fio: Try increasing number of available CPUs\n"); + td_verror(td, EINVAL, "cpus_split"); + goto err; + } + } ret = fio_setaffinity(td->pid, o->cpumask); if (ret == -1) { td_verror(td, errno, "cpu_set_affinity"); @@ -1211,7 +1365,7 @@ static void *thread_main(void *data) #ifdef CONFIG_LIBNUMA /* numa node setup */ if (o->numa_cpumask_set || o->numa_memmask_set) { - int ret; + struct bitmask *mask; if (numa_available() < 0) { td_verror(td, errno, "Does not support NUMA API\n"); @@ -1219,7 +1373,9 @@ static void *thread_main(void *data) } if (o->numa_cpumask_set) { - ret = numa_run_on_node_mask(o->numa_cpunodesmask); + mask = numa_parse_nodestring(o->numa_cpunodes); + ret = numa_run_on_node_mask(mask); + numa_free_nodemask(mask); if (ret == -1) { td_verror(td, errno, \ "numa_run_on_node_mask failed\n"); @@ -1229,12 +1385,16 @@ static void *thread_main(void *data) if (o->numa_memmask_set) { + mask = NULL; + if (o->numa_memnodes) + mask = numa_parse_nodestring(o->numa_memnodes); + switch (o->numa_mem_mode) { case MPOL_INTERLEAVE: - numa_set_interleave_mask(o->numa_memnodesmask); + numa_set_interleave_mask(mask); break; case MPOL_BIND: - numa_set_membind(o->numa_memnodesmask); + numa_set_membind(mask); break; case MPOL_LOCAL: numa_set_localalloc(); @@ -1247,6 +1407,9 @@ static void *thread_main(void *data) break; } + if (mask) + numa_free_nodemask(mask); + } } #endif @@ -1304,6 +1467,9 @@ static void *thread_main(void *data) goto err; } + if (td->flags & TD_F_COMPRESS_LOG) + tp_init(&td->tp_data); + fio_verify_init(td); fio_gettime(&td->epoch, NULL); @@ -1332,22 +1498,28 @@ static void *thread_main(void *data) prune_io_piece_log(td); - verify_bytes = do_io(td); + if (td->o.verify_only && (td_write(td) || td_rw(td))) + verify_bytes = do_dry_run(td); + else + verify_bytes = do_io(td); clear_state = 1; + fio_mutex_down(stat_mutex); if (td_read(td) && td->io_bytes[DDIR_READ]) { - elapsed = utime_since_now(&td->start); + elapsed = mtime_since_now(&td->start); td->ts.runtime[DDIR_READ] += elapsed; } if (td_write(td) && td->io_bytes[DDIR_WRITE]) { - elapsed = utime_since_now(&td->start); + elapsed = mtime_since_now(&td->start); td->ts.runtime[DDIR_WRITE] += elapsed; } if (td_trim(td) && td->io_bytes[DDIR_TRIM]) { - elapsed = utime_since_now(&td->start); + elapsed = mtime_since_now(&td->start); td->ts.runtime[DDIR_TRIM] += elapsed; } + fio_gettime(&td->start, NULL); + fio_mutex_up(stat_mutex); if (td->error || td->terminate) break; @@ -1363,61 +1535,40 @@ static void *thread_main(void *data) do_verify(td, verify_bytes); - td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); + fio_mutex_down(stat_mutex); + td->ts.runtime[DDIR_READ] += mtime_since_now(&td->start); + fio_gettime(&td->start, NULL); + fio_mutex_up(stat_mutex); if (td->error || td->terminate) break; } update_rusage_stat(td); - td->ts.runtime[DDIR_READ] = (td->ts.runtime[DDIR_READ] + 999) / 1000; - td->ts.runtime[DDIR_WRITE] = (td->ts.runtime[DDIR_WRITE] + 999) / 1000; - td->ts.runtime[DDIR_TRIM] = (td->ts.runtime[DDIR_TRIM] + 999) / 1000; td->ts.total_run_time = mtime_since_now(&td->epoch); td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ]; td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE]; td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; - fio_unpin_memory(td); + if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) && + (td->o.verify != VERIFY_NONE && td_write(td))) { + struct all_io_list *state; + size_t sz; - fio_mutex_down(writeout_mutex); - if (td->bw_log) { - if (o->bw_log_file) { - finish_log_named(td, td->bw_log, - o->bw_log_file, "bw"); - } else - finish_log(td, td->bw_log, "bw"); - } - if (td->lat_log) { - if (o->lat_log_file) { - finish_log_named(td, td->lat_log, - o->lat_log_file, "lat"); - } else - finish_log(td, td->lat_log, "lat"); - } - if (td->slat_log) { - if (o->lat_log_file) { - finish_log_named(td, td->slat_log, - o->lat_log_file, "slat"); - } else - finish_log(td, td->slat_log, "slat"); - } - if (td->clat_log) { - if (o->lat_log_file) { - finish_log_named(td, td->clat_log, - o->lat_log_file, "clat"); - } else - finish_log(td, td->clat_log, "clat"); - } - if (td->iops_log) { - if (o->iops_log_file) { - finish_log_named(td, td->iops_log, - o->iops_log_file, "iops"); - } else - finish_log(td, td->iops_log, "iops"); + state = get_all_io_list(td->thread_number, &sz); + if (state) { + __verify_save_state(state, "local"); + free(state); + } } - fio_mutex_up(writeout_mutex); + fio_unpin_memory(td); + + fio_writeout_logs(td); + + if (td->flags & TD_F_COMPRESS_LOG) + tp_exit(&td->tp_data); + if (o->exec_postrun) exec_string(o, o->exec_postrun, (const char *)"postrun"); @@ -1436,11 +1587,12 @@ err: cleanup_io_u(td); close_ioengine(td); cgroup_shutdown(td, &cgroup_mnt); + verify_free_state(td); if (o->cpumask_set) { - int ret = fio_cpuset_exit(&o->cpumask); - - td_verror(td, ret, "fio_cpuset_exit"); + ret = fio_cpuset_exit(&o->cpumask); + if (ret) + td_verror(td, ret, "fio_cpuset_exit"); } /* @@ -1449,10 +1601,17 @@ err: if (o->write_iolog_file) write_iolog_close(td); - fio_mutex_remove(td->rusage_sem); - td->rusage_sem = NULL; + fio_mutex_remove(td->mutex); + td->mutex = NULL; td_set_runstate(td, TD_EXITED); + + /* + * Do this last after setting our runstate to exited, so we + * know that the stat thread is signaled. + */ + check_update_rusage(td); + return (void *) (uintptr_t) td->error; } @@ -1466,7 +1625,7 @@ static int fork_main(int shmid, int offset) struct thread_data *td; void *data, *ret; -#ifndef __hpux +#if !defined(__hpux) && !defined(CONFIG_NO_SHM) data = shmat(shmid, NULL, 0); if (data == (void *) -1) { int __err = errno; @@ -1487,6 +1646,13 @@ static int fork_main(int shmid, int offset) return (int) (uintptr_t) ret; } +static void dump_td_info(struct thread_data *td) +{ + log_err("fio: job '%s' hasn't exited in %lu seconds, it appears to " + "be stuck. Doing forceful exit of this job.\n", td->o.name, + (unsigned long) time_since_now(&td->terminate_time)); +} + /* * Run over the job map and reap the threads that have exited, if any. */ @@ -1564,6 +1730,17 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate, } } + /* + * If the job is stuck, do a forceful timeout of it and + * move on. + */ + if (td->terminate && + time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) { + dump_td_info(td); + td_set_runstate(td, TD_REAPED); + goto reaped; + } + /* * thread is not dead, continue */ @@ -1587,9 +1764,81 @@ reaped: fio_terminate_threads(TERMINATE_ALL); } +static int __check_trigger_file(void) +{ + struct stat sb; + + if (!trigger_file) + return 0; + + if (stat(trigger_file, &sb)) + return 0; + + if (unlink(trigger_file) < 0) + log_err("fio: failed to unlink %s: %s\n", trigger_file, + strerror(errno)); + + return 1; +} + +static int trigger_timedout(void) +{ + if (trigger_timeout) + return time_since_genesis() >= trigger_timeout; + + return 0; +} + +void exec_trigger(const char *cmd) +{ + int ret; + + if (!cmd) + return; + + ret = system(cmd); + if (ret == -1) + log_err("fio: failed executing %s trigger\n", cmd); +} + +void check_trigger_file(void) +{ + if (__check_trigger_file() || trigger_timedout()) { + if (nr_clients) { + if (trigger_remote_cmd) + fio_clients_send_trigger(trigger_remote_cmd); + } else { + verify_save_state(); + fio_terminate_threads(TERMINATE_ALL); + exec_trigger(trigger_cmd); + } + } +} + +static int fio_verify_load_state(struct thread_data *td) +{ + int ret; + + if (!td->o.verify_state) + return 0; + + if (is_backend) { + void *data; + + ret = fio_server_get_verify_state(td->o.name, + td->thread_number - 1, &data); + if (!ret) + verify_convert_assign_state(td, data); + } else + ret = verify_load_state(td, "local"); + + return ret; +} + static void do_usleep(unsigned int usecs) { check_for_running_stats(); + check_trigger_file(); usleep(usecs); } @@ -1599,8 +1848,8 @@ static void do_usleep(unsigned int usecs) static void run_threads(void) { struct thread_data *td; - unsigned long spent; unsigned int i, todo, nr_running, m_rate, t_rate, nr_started; + uint64_t spent; if (fio_gtod_offload && fio_start_gtod_thread()) return; @@ -1629,7 +1878,7 @@ static void run_threads(void) nr_process > 1 ? "es" : ""); } log_info("\n"); - fflush(stdout); + log_info_flush(); } todo = thread_number; @@ -1643,12 +1892,16 @@ static void run_threads(void) if (!td->o.create_serialize) continue; + if (fio_verify_load_state(td)) + goto reap; + /* * do file setup here so it happens sequentially, * we don't want X number of threads getting their * client data interspersed on disk */ if (setup_files(td)) { +reap: exit_value++; if (td->error) log_err("fio: pid=%d, err=%d/%s\n", @@ -1698,9 +1951,9 @@ static void run_threads(void) } if (td->o.start_delay) { - spent = mtime_since_genesis(); + spent = utime_since_genesis(); - if (td->o.start_delay * 1000 > spent) + if (td->o.start_delay > spent) continue; } @@ -1836,57 +2089,70 @@ static void run_threads(void) update_io_ticks(); } -void wait_for_disk_thread_exit(void) +static void wait_for_helper_thread_exit(void) { - fio_mutex_down(disk_thread_mutex); + void *ret; + + helper_exit = 1; + pthread_cond_signal(&helper_cond); + pthread_join(helper_thread, &ret); } static void free_disk_util(void) { - disk_util_start_exit(); - wait_for_disk_thread_exit(); disk_util_prune_entries(); + + pthread_cond_destroy(&helper_cond); } -static void *disk_thread_main(void *data) +static void *helper_thread_main(void *data) { int ret = 0; fio_mutex_up(startup_mutex); - while (threads && !ret) { - usleep(DISK_UTIL_MSEC * 1000); - if (!threads) - break; + while (!ret) { + uint64_t sec = DISK_UTIL_MSEC / 1000; + uint64_t nsec = (DISK_UTIL_MSEC % 1000) * 1000000; + struct timespec ts; + struct timeval tv; + + gettimeofday(&tv, NULL); + ts.tv_sec = tv.tv_sec + sec; + ts.tv_nsec = (tv.tv_usec * 1000) + nsec; + if (ts.tv_nsec > 1000000000ULL) { + ts.tv_nsec -= 1000000000ULL; + ts.tv_sec++; + } + + pthread_cond_timedwait(&helper_cond, &helper_lock, &ts); + ret = update_io_ticks(); + if (helper_do_stat) { + helper_do_stat = 0; + __show_running_run_stats(); + } + if (!is_backend) print_thread_status(); } - fio_mutex_up(disk_thread_mutex); return NULL; } -static int create_disk_util_thread(void) +static int create_helper_thread(void) { int ret; setup_disk_util(); - disk_thread_mutex = fio_mutex_init(FIO_MUTEX_LOCKED); - - ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL); - if (ret) { - fio_mutex_remove(disk_thread_mutex); - log_err("Can't create disk util thread: %s\n", strerror(ret)); - return 1; - } + pthread_cond_init(&helper_cond, NULL); + pthread_mutex_init(&helper_lock, NULL); - ret = pthread_detach(disk_util_thread); + ret = pthread_create(&helper_thread, NULL, helper_thread_main, NULL); if (ret) { - fio_mutex_remove(disk_thread_mutex); - log_err("Can't detatch disk util thread: %s\n", strerror(ret)); + log_err("Can't create helper thread: %s\n", strerror(ret)); return 1; } @@ -1911,40 +2177,49 @@ int fio_backend(void) return 0; if (write_bw_log) { - setup_log(&agg_io_log[DDIR_READ], 0, IO_LOG_TYPE_BW); - setup_log(&agg_io_log[DDIR_WRITE], 0, IO_LOG_TYPE_BW); - setup_log(&agg_io_log[DDIR_TRIM], 0, IO_LOG_TYPE_BW); + struct log_params p = { + .log_type = IO_LOG_TYPE_BW, + }; + + setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log"); + setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log"); + setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log"); } startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED); if (startup_mutex == NULL) return 1; - writeout_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED); - if (writeout_mutex == NULL) - return 1; set_genesis_time(); stat_init(); - create_disk_util_thread(); + create_helper_thread(); cgroup_list = smalloc(sizeof(*cgroup_list)); INIT_FLIST_HEAD(cgroup_list); run_threads(); + wait_for_helper_thread_exit(); + if (!fio_abort) { - show_run_stats(); + __show_run_stats(); if (write_bw_log) { - __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); - __finish_log(agg_io_log[DDIR_WRITE], - "agg-write_bw.log"); - __finish_log(agg_io_log[DDIR_TRIM], - "agg-write_bw.log"); + for (i = 0; i < DDIR_RWDIR_CNT; i++) { + struct io_log *log = agg_io_log[i]; + + flush_log(log); + free_log(log); + } } } - for_each_td(td, i) + for_each_td(td, i) { fio_options_free(td); + if (td->rusage_sem) { + fio_mutex_remove(td->rusage_sem); + td->rusage_sem = NULL; + } + } free_disk_util(); cgroup_kill(cgroup_list); @@ -1952,8 +2227,6 @@ int fio_backend(void) sfree(cgroup_mnt); fio_mutex_remove(startup_mutex); - fio_mutex_remove(writeout_mutex); - fio_mutex_remove(disk_thread_mutex); stat_exit(); return exit_value; }